Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Tropical Cyclone Modeling and Data Assimilation
Jason Sippel NOAA AOML/HRD
2018 WMO Workshop at NHC
• History of TC forecast improvements in relation to model development
• Ongoing modeling/DA developments
• Future direction
Outline
History of improvements: Error trends
• Hurricane track forecasts have improved markedly
• The average Day-3 forecast location error is now about what Day-1 error was in 1990
• These improvements are largely tied to improvements in large-scale forecasts
Hurricane Location Forecast Errors: 1990-2017
1990 2017
300
100
History of improvements: Error trends
• Hurricane track forecasts have improved markedly
• The average Day-3 forecast location error is now about what Day-1 error was in 1990
• These improvements are largely tied to improvements in large-scale forecasts
History of improvements: Error trends
• Hurricane intensity forecasts have only recently improved
• Improvement in intensity forecast largely corresponds with commencement of Hurricane Forecast Improvement Project
Hurricane Intensity Forecast Errors: 1990-2017
1990 2017
20
10
30
HFIP era
History of improvements:Error trends
• Significant focus of HFIP has been the development of the HWRF model
• As a result, HWRF intensity errors have decreased significantly over the past decade
HWRF intensity error at 72 h has decreased by almost 50% in last decade!
History of improvements:Using TC Observations
• US has used dropsondes for TC model forecast improvement since 1997
• Aberson (2010, 2011) examined impact of dropsondes in GFS
• Significant track improvement globally
Percent improvement as a result of assimilating NOAA, DOTSTAR, and THORPEX dropsondes in September 2008 (Aberson 2011)
History of improvements:Using TC Observations
• Recent GFS (v2017) retrospectives assimilated Global Hawk dropsondes
• SUBSTANTIAL benefits for GFS track!!!
• Ongoing work suggests very high altitude of GH sondes is important
Track error – 2016 NATL
Track skill – 2016 NATL
History of improvements:Using TC Observations
• Starting in 2008, it became apparent that assimilating Doppler velocity data had potential for forecast improvement
• Assimilating radar data significantly improved analyses and forecasts of Hurricane HumbertoObservations (left) and analyses (right) of reflectivity
from Hurricane Humberto with an experimental system
History of improvements:Using TC Observations
• Starting in 2008, it became apparent that assimilating Doppler velocity data had potential for forecast improvement
• Assimilating radar data significantly improved analyses and forecasts of Hurricane Humberto
Obs and NODA forecasts
Obs and DA forecasts
History of improvements:Using TC Observations
• Subsequent work showed forecast improvements from assimilating Doppler velocity from recon (TDR)
• For the sample examined, the forecast improvement was significant
Operational and experimental intensity forecasts of Hurricane Ike (2008) prior to landfall near Houston. The forecast from EnKF used assimilation of TDR velocity data.
History of improvements:Using TC Observations
• Subsequent work showed forecast improvements from assimilating Doppler velocity from recon (TDR)
• For the sample examined, the forecast improvement was significant
Operational and experimental intensity forecasts of cases from 2008-2010. The forecast from EnKF used assimilation of TDR velocity data.
History of improvements:Using TC Observations
• As a result of this work, TDR data began being assimilated in HWRF in 2013
• For weak storms like Karen, there was substantial improvement of a positive intensity bias in HWRF (purple)
Forecast of Karen from HWRF without(top) and with (bottom) TDR DA
• Results diminished over a larger sample (cf red & black lines)
• Major problem was a substantial negative bias in the first 24 h
• The problem is worse for stronger storms and is the result of physics and DA deficiencies
History of improvements:Using TC Observations
Consistent Vmax benefit for TS cycles
Spindown for strong storms due to inaccurate error covariance and physics deficiencies
History of improvements:Using TC Observations
• HDOB data (i.e., flight-level and SFMR) reduce track error more (more data continuity)
• Spindown problem worse for HDOB data
• Best results from using all data
History of improvements:Using TC Observations
CURRENT OBSERVATIONS ASSIMILATED BY HWRF INCLUDE:
• Conventional observations (radiosondes, dropwindsondes, aircraft reports, ship and buoy obs, surface observations over land, pibal winds, wind profilers, VAD wind, WindSatscatterometer wins, PW derived from GPS)
• NOAA P3 aircraft Tail Doppler Radar
• Flight-level data from reconnaissance
• IR/VIS cloud drift winds and water vapor winds from GOES, EUMETSAT, MODIS, JMA
• Clear-sky satellite radiance observations including data from HIRS, AIRS, IASI, GOES Sounders, CrIS, SSMIS, Metop-B, AMSU-A, MHS and IASI
History of improvements:Using TC Observations
History of improvements:Battling spindown
• Vortex initialization (VI) procedure in HWRF produces less spindownthan does DA
• Blending zero’s DA increments near center for Vmax > 64 kt
• Not a long-term solution
History of improvements:Battling spindown
• Recent work showed that increasing resolution AND improving physics (diffusion/mixing) are necessary to reduce spindown
• The challenge is to make physics changes that don’t make every TD a Cat 5
Experimental HWRF forecasts of RI ofHurricane Patricia
History of improvements:Battling spindown
• It was found that unrealistic wind profiles were causing DA problems in HWRF
• Lowering the drag coefficient produces better wind profiles and improves DA
• Other PBL changes have been made, more neededOld (top) and new (bottom) HWRF wind
profiles as a result of changing Cd
History of improvements:Battling spindown
• It also was evident that DA system improvements were necessary to reduce spindown
• Results from experimental OU system showed significant improvements with use of self-cycled covariance
Vmax errors for Hurricane Edouard in the operational HWRF vs the experimental OU HWRF system with fully-cycled covariance.
History of improvements:Battling spindown
History of improvements:Battling spindown
• New DA system shows less negative bias during early part of forecast
• To this point this does not improve intensity error
• Ongoing tuning is improving performance
Intensity bias from the old and new HWRD data assimilation systems
• History of TC forecast improvements in relation to model development
• Ongoing modeling/DA developments
• Future direction
Outline
Ongoing developments:Using more observations
• Dropsonde observations currently transmitted in TEMPDROP format
• Only report release location in main body
• NCEP does not consider drift, which causes problems in the vortex
Splash location
Release
location
Max (111 km, 3*RMW)
x
x
x
x
x
x
x
Ongoing developments:Using more observations
• U/V rejected in vortex due to concerns regarding drift
• Current rejection radius is max(111 km, 3*RMW)
• This gets rid of a lot of data (too much?)
Ongoing developments:Using more observations
• Drift probably not an issue in all cases
• Unflagging u/v outside of R64 increases intensity skill by 5-10% (neutral track)
• Suggests we should be dropping far more in vortex
Change in intensity skill due to assimilating more dropsonde winds in the TC vortex
Ongoing developments:Using more observations
• Code has been developed to estimate dropsonde location
• Ongoing test uses all dropsonde observations with estimated location
• Track improves up to 5%, intensity neutral (different sample)
Change in track skill due to considering dropsonde drift and using all data
Ongoing developments:Physics tests
• Current HWRF PBL is GFS PBL + band aids + band aids
• Wholesale change likely needed, and YSU is the best candidate
• Initial tests look very promising for both track and intensity!
Change in track skill due to using YSU PBL
• History of TC forecast improvements in relation to model development
• Ongoing modeling/DA developments
• Future direction
Outline
Courtesy Xuguang Wang, HFIP partner
High-frequency full cycling alleviates imbalance.
3DEnVAR – 6h 3DEnVAR – 1h
Considering rapid error evolution reduces imbalance
A comparison of the HRD TDR analysis with the OU 3DEnVar hybrid HWRF analysis from Edouard with 6-h and 1-h cycling.
Future direction: DA
Courtesy Xuguang Wang, HFIP partner
4DEnVAR alleviates imbalance as well.
3DEnVAR – 6h 4DEnVAR – 6h
Considering rapid error evolution reduces imbalance
A comparison of the HRD TDR analysis with the OU 3DEnVar hybrid HWRF analysis from Edouard with 6-h and 1-h cycling.
Future direction: DA
On average, forecasts initialized from 4DEnVAR have lower Vmax error for first 36h
Courtesy Xuguang Wang, HFIP partner
Track, SLP, and Vmax errors from Edouard in the OU hybrid 3DVar and 4DVar systems
Future direction: DA
Vmax from the 12Z17 cycle of Edouard in the OU hybrid 3DVar and 4DVar systems. Courtesy Xuguang Wang, HFIP partner.
Future direction: DA
• Results from OU system show hourly cycling helps with inner core balance
• Current priority is to develop/test this for operational HWRF
• This should appeal to researchers as well
Future direction: Other DA issues
• Replace vortex initialization with self-consistent DA of something derived from TCVitals
• Update condensate (and w?) with each cycle
• Assimilation of cloudy radiances
• Coupled atmosphere-ocean DA
Vertical cross-section of U analysisincrements (filled) and U wind(contoured) in a cycle of Irma. Notestrong anticyclonic increments.
Future direction: Other non-DA issues
• Intensity-dependent biases (overintensificationof weak systems)
• Multi-storm approach (e.g., basin-scale)
• Probabilistic forecasting
• Targeting
Track skill of 2017 basin-scale HWRF as compared with operational HWRF and GFS
Conclusions
• NOAA TC prediction is undergoing dramatic advancements, lead by improvements in global models and HWRF
• We are using more of the available data in DA
• Long term plans address ongoing issues (e.g., spindown, bias) and allow for greater data usage
• The above factors should contribute to intensity improvement in particular